hermes - 💡(How to fix) Fix scripts/run_tests.sh leaks HERMES_CRON_SESSION into pytest under cron [5 pull requests]

Bug Description

The canonical test runner claims to scrub behavioral HERMES_* variables, but it does not unset HERMES_CRON_SESSION. When the test runner is invoked from a scheduled Hermes cron job, that environment variable leaks into the pytest process and changes approval behavior from the normal non-interactive auto-approve path to cron-deny behavior.

This makes otherwise passing approval tests fail only under cron/scheduled bug-hunt runs, and can hide real regressions behind environment contamination.

Reproduction

From current main at f2afa68a4:

# Simulate the cron environment leaking into pytest
HERMES_CRON_SESSION=1 venv/bin/python -m pytest -o addopts= -n 0 \
  tests/acp/test_approval_isolation.py::TestAcpExecAskGate::test_interactive_env_var_routes_to_callback \
  -vv --tb=short

Observed failure:

E       assert False is True
FAILED tests/acp/test_approval_isolation.py::TestAcpExecAskGate::test_interactive_env_var_routes_to_callback

Control run passes when the variable is removed:

env -u HERMES_CRON_SESSION venv/bin/python -m pytest -o addopts= -n 0 \
  tests/acp/test_approval_isolation.py::TestAcpExecAskGate::test_interactive_env_var_routes_to_callback \
  -q
# 1 passed in 2.27s

Expected Behavior

scripts/run_tests.sh should fully hermeticize the test environment it documents. Running it from a Hermes cron job should not make tests observe cron-only approval mode.

Actual Behavior

scripts/run_tests.sh unsets many behavioral variables (HERMES_INTERACTIVE, HERMES_GATEWAY_SESSION, HERMES_EXEC_ASK, etc.) but not HERMES_CRON_SESSION, so cron approval mode leaks into tests.

Relevant file:

• scripts/run_tests.sh, lines 65-73: behavioral var scrub list omits HERMES_CRON_SESSION.
• tools/approval.py, lines 865-879: HERMES_CRON_SESSION changes dangerous-command approval behavior.

Suggested Fix

Add HERMES_CRON_SESSION to the scrub list in scripts/run_tests.sh. It may also be worth adding a small regression test or documenting the complete set of scheduler-injected variables.

Environment

• Repo: NousResearch/hermes-agent
• Branch: main
• Commit tested: f2afa68a4
• Runner context: Hermes scheduled cron job on WSL/Linux
• Python: venv/bin/python 3.11.15